Articleshttps://doi.org/10.1038/s41561-019-0505-5

1GFZ German Research Centre for Geosciences, Potsdam, Germany. 2Observatoire Midi Pyrénées, IRAP, CNRS UMR 5277, Université Paul Sabatier, Toulouse, France. 3University of Kiel, Kiel, Germany. 4King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia. 5University of Potsdam, Potsdam, Germany. *e-mail: simone.cesca@gfz-potsdam.de

In May–June 2018, global earthquake monitoring agencies detected a series of large earthquakes in an otherwise relatively quiet area ~35 km east of the island of Mayotte. A peak MW 5.91

earthquake occurred on 15 May 2018, the largest ever recorded in the region. In November 2018, long-duration (~20 min) very-long-period seismic signals (VLPs) with periods of ~16 s were discov-ered in global seismic recordings1, triggering the curiosity of the scientific community. The origin of the VLPs was traced back in the vicinity of the swarm activity close to Mayotte. Most often, VLP sig-nals2 are recorded in volcanic areas3 suggesting a magmatic origin for the sequence.

Mayotte is one of the four principal volcanic islands in the Comoros archipelago and home to a population of ~256,000 (2017). The island belongs to a 250 km long NW–SE chain of basaltic volca-noes located between Africa and Madagascar (Fig. 1). The region has been affected by multiple tectonic processes4,5 including an episode of NE–SW trend rifting during the Permo–Triassic, associated with the fragmentation of Gondwana6,7 and the formation of the Somali and Mozambique oceanic basins, during which Madagascar drifted southwards8–12. Proposed sources of volcanism include hotspot13, passive magma ascent through lithospheric discontinuities14 or rift-ing coeval to rifting in Southern East Africa15. Volcanism at Mayotte started at about 10–20 Ma (ref. 15,16) and subsequently migrated to produce the other islands. Mayotte last erupted 4,000 ± 500 years ago15. The nature of the crust and its thickness beneath the Comoro Islands is debated14,15. Only a few M > 4 earthquakes have been recorded in this area, including the 1993 Mb 5.2 event, which caused ~1.7 M Euros of damage in Mayotte, and the 2011 Mb 4.9 (ref. 16). Focal mechanisms and GNSS data support NE–SW transtension17,18.

By modelling seismological data at regional and teleseis-mic distances and ground displacement recordings at Mayotte (Supplementary Fig. 1) collected between May 2018 and April 2019,

we provide evidence of the drainage of 1.7 ± 0.4 km3 of magma from a ~30 km deep sub-Moho magma reservoir by a dyke that propagated to the seafloor. The dyke propagation caused a swarm of almost 7,000 volcano-tectonic earthquakes (VTs) and the downsag of the host rock overlying the reservoir, which in turn triggered its resonance, emitting 407 long-duration VLPs, and helped sustain a high outflow rate throughout. We argue that reservoir roof failures at calderas, which are well-studied, can be used as a downscaled analogue to evaluate future scenarios.

Data analysis and modellingWe use seismic data at regional and teleseismic distances (ABKAR seismic array, Kazakhstan) and test different velocity models (Supplementary Fig. 2, Supplementary Tables 1–4) to perform full waveform moment tensor (MT) inversions and depth phase analysis, thereby retrieving focal mechanisms, centroid locations (Supplementary Figs. 3–5) and depths (Supplementary Figs. 6–8) for the most energetic VTs19. Additionally, we take advantage of seismic station YTMZ—deployed over the full study period on Mayotte—to produce an enhanced catalogue19 of relative locations of weaker VTs (Supplementary Fig. 9). We detected 6,990 VTs and located 1,904 of them (Fig. 1) by retrieving their origin direction and distance (Supplementary Figs. 10 and 11; see Supplementary Information). The VTs are classified into families of events (colour-coded in Figs. 1 and 2a–c) with similar waveforms and distance to station YTMZ (Supplementary Fig. 12). We design a detection tool to scan the broadband data at four regional seismic stations for monochro-matic, low-frequency signals, thereby detecting 407 long-duration VLPs19 (Supplementary Fig. 13). We also develop an algorithm to invert for the centroid MT of the VLPs and resolve the damping constant and dominant frequency of a damped linear oscillator as the source time function, obtaining high-quality solutions for

Drainage of a deep magma reservoir near Mayotte inferred from seismicity and deformationSimone Cesca   1*, Jean Letort2, Hoby N. T. Razafindrakoto   1, Sebastian Heimann1, Eleonora Rivalta   1, Marius P. Isken3, Mehdi Nikkhoo1, Luigi Passarelli   4, Gesa M. Petersen   1,5, Fabrice Cotton1,5 and Torsten Dahm1,5

The dynamics of magma deep in the Earth’s crust are difficult to capture by geophysical monitoring. Since May 2018, a seis-mically quiet area offshore of Mayotte in the western Indian Ocean has been affected by complex seismic activity, including long-duration, very-long-period signals detected globally. Global Navigation Satellite System stations on Mayotte have also recorded a large surface deflation offshore. Here we analyse regional and global seismic and deformation data to provide a one-year-long detailed picture of a deep, rare magmatic process. We identify about 7,000 volcano-tectonic earthquakes and 407 very-long-period seismic signals. Early earthquakes migrated upward in response to a magmatic dyke propagating from Moho depth to the surface, whereas later events marked the progressive failure of the roof of a magma reservoir, triggering its resonance. An analysis of the very-long-period seismicity and deformation suggests that at least 1.3 km3 of magma drained from a reservoir of 10 to 15 km diameter at 25 to 35 km depth. We demonstrate that such deep offshore magmatic activity can be captured without any on-site monitoring.

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience 87

Articles Nature GeoscieNce

22 VLPs19 (Fig. 3). Due to the uncertainties on Moho depth and crustal structure, all our results are tested against different velocity models with varying Moho depth (see Supplementary Information).

Chronology of the crisisWe have identified four phases of the volcano-seismic crisis. During Phase I (10 May to 7 June 2018) the most energetic VT burst of the sequence (11 MW > 5 events between 15 and 20 May) occurred ~35 km East of Mayotte (Fig. 2a). We find a depth of 22 km for the largest MW 5.9 earthquake. Between 31 May and 7 June 2018, the seismicity migrated upward, as confirmed by the centroid depth estimations and array beam analysis (Fig. 2c, Supplementary Figs. 7 and 8). During Phase II (7 June to 18 June 2018) epicentres migrated southwards approaching the recently discovered seamount20 (Fig. 1, Supplementary Fig. 5). Left-lateral strike-slip faulting dominates in Phases I–II (Fig. 1). A consistent expansion-related isotropic com-ponent, increasing gradually up to 20% of the total moment, is found during the upward migration and in Phase II (Supplementary Figs. 5 and 6). By mid June, a new type of activity emerged in the form of

long-duration VLPs. However, early VLPs had already occurred on 30 January and 2 June.

Few Mw > 4.5 VTs (Fig. 2a) occurred in Phase III (28 June to 17 September 2018), while VLPs became dominant. Phase IV (17 September 2018 to March 2019) started with an increasing VLP rate, this time accompanied by a new type of VTs (Fig. 2c,d). The dominant period of VLPs increased smoothly from ~15.2 s (June 2018) to a maximum of ~15.6 s (October 2018), before decreasing again to ~15.3 s (February 2019; Fig. 2e, Supplementary Fig. 13).

The VLP MTs are similar (Fig. 1), with a predominance of alter-nating positive and negative vertical Compensated Linear Vector Dipole (CLVD) (Fig. 3). The vertically axisymmetric MTs suggest a subhorizontal crack, dipping slightly to the west. Centroid locations are the same within errors and depths are at 37 ± 11 km. The largest VLP occurred on 11 November 2018, with an estimated surface wave magnitude of MS 5.1. The damping of the VLP source (quality fac-tor, Q) is in the range 72 ± 6, which is higher than in other cases21,22 but not unique23. VLPs onsets are often quasi-simultaneous to one or multiple weak VTs, as observed previously23. Due to the emergent

50 40 30 20 10

Depth (km)La

titud

e (°

S)

Longitude (° E)

Latitude (° S)

45.1 45.2

GAMO KAWEYTMZ

Petite Terre

Mw 5.9

MAYGMamoudzou

BDRL

Grand Camore

45.3 45.4 45.5 45.6

45.4 45.5

?

?

45.7 45.8 45.9

45.6 45.7 45.8 45.8

12.6

12.7

12.8

12.9

13.0

12.6

12.7

12.8

12.9

13.0

10

20

Average uncertainty

10 km

10 cm

Centroid-moment-tensor depth

Depth (km

)

12

Latit

ude

(° S

)

b

a

Mayotte

Davie Fracture

Zone

Mozambique channel Mad

agas

car13

14

42 44 46 48 Longitude (° E)Longitude (° E)

30

40

50

Fig. 1 | Map view and cross sections of seismic and deformation sources. a, Weak VT locations (points) and strong VT MTs (focal sphere, lower emisphere projections of the double-couple components in map view and backprojection of full MTs from East or South in the cross sections, respectively) plotted and colour-coded according to similarity of waveform and distance to station yTMZ (yellow triangle), overlay of 22 VLPs MTs (black focal spheres, negative CLVD convention used), best-fit point deformation sources (black symbols, see Supplementary Information), vertical displacements (black lines) and time evolution of horizontal displacements at four GNSS stations (black scatter points, 1 July 2018 to 1 April 2019); uncertainties are within 2σ of vertical displacements (green boxes, horizontal uncertainties are negligible with respect to vertical ones), best fitting modelled displacements (red lines and arrows) and the inferred VLP source location (dashed red ellipses). The southeast and upward migration paths of seismicity and magma in Phases I–II (dashed black arrows in the cross-sections) reach the location of the discovered seamount20 (orange circle and bars, assuming a 5 km diameter20). b, Map of the Comoro Islands, major regional tectonic structures and basin configuration5, showing fracture zone (thin lines) and earthquake locations, and MTs before the sequence (circle and focal spheres, source GlobalCMT, GEOFON and USGS catalogues); shaded-relief topography and bathymetry are from the ETOPO1 Global Relief Model from NOAA’s National Centers for Environmental Information. The location of the study region is highlighted by a red star on the Earth globe.

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience88

ArticlesNature GeoscieNce

nature of VLP signals, it is difficult to judge whether VTs precede or follow VLPs, but some VLPs respond to VTs with abrupt signal amplitude changes, depending on whether the successive VTs are in or out of phase with the resonating VLP source (Fig. 4a). This obser-vation suggests that VTs act as repeated forcing, contributing to the exceptionally long duration of some VLPs. The VLP spectra (Fig. 4b) reveal higher modes including non-integer ratios between them.

All VTs accompanying VLPs and most VTs in Phases III–IV (red and green dots, Figs. 1 and 2a,c) are located closer to Mayotte and have steep NE–SW striking thrust mechanisms (Fig. 1), inconsistent with local transtension. An isolated burst of VTs at the end of

August 2018 included earthquake repeaters and anti-repeaters, with highly correlated and anti-correlated waveforms, respectively (Fig. 4c). VTs and VLPs have persisted throughout Phase IV until the time of writing. The cumulative moment of VTs in Phases III–IV is M0 = 5.42·1016 Nm, corresponding to Mw = 5.1, far less than for the dyke-related seismicity in Phases I–II (M0 = 2.03·1018 Nm, Mw = 6.2).

Geodetic data from Mayotte Island reveal a steady, long-lasting subsidence and eastward displacement at four GNSS stations. The length of the displacement vector from July 2018 through April 2019 is 18 cm. Early ground deformation is weak, and grows clearly from July onward. We constrain the location, depth, aspect ratio and

6 VTa

b

c

d

e

f

g

VT

VT

VLP

VLP

Deformation

Dyke VTs ( )

Sagging VTs ( )

VLPs ( )

Deformation ( )

Upwardmigration

5

4

0

10

20

30

8

6

4

5

4

3

15.6

15.4

15.2

North

10.0

7.5

5.0

2.5

0

200

150

100

50

0

Absolute

displacement (m

m)

Daily rate

40

20

0

Daily rate

Vertical

East

0

–100

100

Anomalyonset

Nor

mal

ized

no. o

f VT

/VLP

Dis

plac

emen

t(m

m)

Per

iod

(s)

Ms

t S–t

P (

s)D

epth

(km

)M

W

1.0

0.5

0

Date

June

201

8

July

2018

Augus

t 201

8

Septe

mbe

r 201

8

Octobe

r 201

8

Novem

ber 2

018

Decem

ber 2

018

Janu

ary 2

019

Febr

uary

201

9

Mar

ch 2

019

Fig. 2 | Timeline of the seismic sequence. a, VT magnitudes (colour-coding as in Fig. 1) and seismicity rate (black line). b, VT depth based on MT inversion (circles) and array analysis (crosses). c, Differential S–P time of VTs at station yTMZ as a proxy of the distance to Mayotte (see Supplementary Information for uncertainties). d, VLP magnitudes (Ms, purple circles) and rate (black line). e, VLP dominant periods (purple circles). f, Demeaned and detrended East, North and vertical GNSS displacements at station MAyG. g, Normalized cumulative number of dyke-related VTs (the blue line refers to blue, cyan and purple VTs in a–c), sagging-related VTs (the red line refers to red and green VTs in a–c), VLPs (indigo line) and length of GNSS displacement vectors at station MAyG (green line). Sequence phases are marked in all panels with cyan vertical bars.

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience 89

Articles Nature GeoscieNce

volume change by inverting these GNSS data for the time interval 1 July 2018 to 31 March 2019. An isotropic source of pressure is unable to simultaneously fit the ratio of horizontal to vertical displace-ment and the vectors’ orientation1. We use generalized point-source deformation models24,25, first assuming a vertical ellipsoidal source26 and investigating the effect of layering on the inferred source shape (Supplementary Figs. 14 and 15). The best fit is given by a large nega-tive isotropic plus a small positive vertical CLVD source 12 ± 4 km East of Petit Terre at a depth of 32 ± 3 km, volume change is here –1.7 ± 0.4 km3. We also test non-axisymmetric generalized point sources in a homogenous medium24, obtaining a laterally contract-ing vertical dislocation as the best-fit deformation mechanism at a

depth of 23 km and expelled magma volume (potency) of −1.3 km3 with 99% confidence bounds of (20, 27) km and (−2.8, −1.2) km3, respectively. This is a lower bound of the source potency since all GNSS stations are grouped on one side of the source, so that the net-work has no sensitivity to NS source contraction. Both analyses indi-cate that vertical shrinking of the source is negligible, suggesting that the shrinking source does not coincide with the source of the VLPs.

Interpretation of chain of eventsIn summary, we have identified two main stages of the unrest (Fig. 5). The first stage (Phases I–II), with the migration of an energetic seismic swarm from deep to shallow depth, is consistent

KIBK.T1,416 km–36°

KIBK.Z1,416 KM–36°

SBV.Z489 km99°

02:37 min 02:38 min 00:46 min

00:46 min

01:13 min

01:13 min

02:38 min

02:39 min

02:39 min

02:37 min

03:00 min

03:00 min

∆15:23 min

∆15:23 min

∆16:07 min

∆16:07 min

∆15:24 min

∆15:24 min

∆15:26 min

∆15:26 min

∆11:06 min

∆11:06 min

∆12:08 min

∆12:08 min

CAM.Z1,408 km131°

CAM.T1,408 km131°

a

c

b

KMBO.T1,578 km–35°

KMBO.Z1,578 km–35°

MSEY.T1,422 km51°

MSEY.Z1,422 km51°

ABPO.T713 km165°

ABPO.Z713 km165°

SBV.T489 km99°

Time (s)

0 100 200 300 400 500 600 700

TRTT = 15.4 s

Fig. 3 | MT solution for the 31 July 2018 VLP. a, The MT plot overlays the focal spheres of the ensemble of bootstrap solutions (see Supplementary Information). The red line denotes the overall best solution. The negative CLVD convention is applied. b, The resolved, oscillating source time function, with a dominant period TR = 15.4 s. c, Comparison of observed (red lines) and synthetic displacement traces (black lines), as well as their differences (red areas) for a selection of stations and components. Synthetic traces were computed assuming a continental (model P2017, see Supplementary Information) or oceanic crust, based on station locations (station name, spatial component, epicentral distance, azimuth, time window starting time, respect to the origin time, and time window duration are reported for each subplot).

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience90

ArticlesNature GeoscieNce

with the propagation of a magmatic dyke through the whole crust. The second stage (Phases III–IV), with simultaneous VLP and VT activity and large deflation, suggests the evacuation of a sub-Moho, large reservoir and a multifaceted interaction between magma and host rock.

The energetic seismic swarms in Phase I marks, to our knowledge, the first case of vertical propagation of a dyke traced for >25 km from a deep magma reservoir to the surface, although dykes have been observed to propagate horizontally for longer distances27,28. Dyke-induced swarms of this magnitude are rare and attributed to large intruded volumes, which scale with the moment of induced events as a power law29. According to this model, the cumulative moment of Phases I–II, M0 = 2·1018 Nm, would correspond to a dyke volume of 5·10−2 km3. Strike-slip focal mechanisms have been observed for other propagating dykes28,30–33. Phase II involves a lat-eral dyke propagation, with VTs migrating 10–20 km towards South and reaching the seamount location20. The relative VT quiescence

in Phase III is consistent with the establishment of an open path-way to the surface. High-rate crustal deformation sets on a week after the beginning of Phase III, consistent with the time needed for magma to widen its pathway by conduit erosion and establish a high flow rate34. Taken together, these observations suggest that the end of Phase II may mark the onset of the submarine eruption. During Phase IV, VTs, VLPs and ground deformation appear interlinked mechanically. VTs plausibly trigger VLPs and accompany the slow evacuation of the magma reservoir (Fig. 2d,f).

VLPs may have different causes, but ringing events such as those observed at Mayotte have previously been explained by the resonance caused by slow standing waves trapped at the fluid–solid interface of a fluid-filled crack or conduit2,35,36. Here, their period and duration are especially long, which may be explained by the large size of the crack20, the stronger host rock because of the depth and basalt as the fluid filling the crack3. We use analyti-cal formulas37 to constrain crack geometry based on the dominant

10:25:00

a

b

c

101

10–2 10–1

Frequency (Hz)

100

100

Ver

tical

acc

eler

atio

nN

orm

aliz

ed a

ccel

erat

ion

Spe

ctra

l rat

io V

LP/n

oise

Time

Time

10:55:00

Low frequencies, BP 0.06–0.08 HzHigh frequencies, BP 0.06–0.50 Hz

Vertical

North

East

23:34:2523:34:15

Fig. 4 | Example of seismic signals and spectra recorded at the YTMZ station. a, 11 November 2018 VLP. Waveforms correspond to normalized vertical acceleration filtered in different frequency bands. b, Stack of normalized vertical spectra for 51 VLPs (indigo bars represent integer multiples of the dominant frequency (0.065 Hz, or 15.4 s) with a red bar for the 2.5 multiple). c, Anti-correlated waveforms for two VTs occurring 1.5 min apart on 20 August 2018 (time windows starting at 21:34:15.5 and 21:35:39.6 respectively). Waveforms of the first event (indigo) are compared to flipped waveforms of the second event (blue).

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience 91

Articles Nature GeoscieNce

VLP period, which we assume is the fundamental resonance mode. Fundamental periods of 15.2–15.6 s are possible for a wide range of lengths and thicknesses (Fig. 6). However, cracks with L < 8 km have too little volume to expel more than 1.5 km3 of magma. A crack with L > 15 km is unlikely, given the distribution of seismicity (Fig. 1). Smooth dominant period changes from 15.2 to 15.6 s and back may result first from the crack predominantly thinning and next shortening (Fig. 6). For example, path for L = 12 km carries a volume loss of 1.6 km3. Thus, the change of dominant period may be entirely explained in terms of geometry changes consistent with the observed volume loss; variations in other parameters such as melt compressional wave velocity are also possible.

VLPs in combination with VTs have often been observed before and during caldera collapses18,38–42. For example, VLPs occurred before and during the 2000 caldera collapse at Miyakejima, Japan, both with oscillating signals39 and 20–50 s single pulse source time functions38,40. With a maximum duration of ~60 s they were consid-erably shorter than those at Mayotte; their modelled source mecha-nism also had a different geometry38–40. They were interpreted as the resonance of an axially symmetric structure in response to a shallower trigger39 or as resulting from the intermittent sinking of a vertical piston into the magmatic chamber, causing its sudden volu-metric change38,40. At Piton de la Fournaise, La Reunion island, 0.02–0.50 Hz VLPs have been attributed to repeated piston-like collapse42. The thrust mechanisms of typical VTs in Phase IV, inconsistent with the NE–SW transtensional background stress regime, require a strong stress perturbation, probably provided by the evacuation of the magma reservoir. Steep outward dipping faults are typically formed in the early stage of reservoir roof failure upon depletion43,44. They occur on newly formed, distributed faults (as suggested by their small magnitude and existence of several families), rather than on a ring fault, confirming that the faulting is weakening the overly-ing material rather than representing the slip of a coherent block. As they are close above the reservoir, they may exert an efficient pressure pulse and trigger waves at the fluid–solid interface travel-ling towards the opposite edge of the reservoir and back, thereby producing a ringing signal3. The observed anti-correlated VT pairs (Fig. 4c) may be explained by the reverse motion on a fault, similar

to observations during collapse events at calderas45. The January 2018 VLP, observed before the dyke had formed, suggests that the oscillator is a portion of the deep reservoir rather than the feeder dyke, and that the reservoir then had a similar size and shape.

Our deformation models locate a volume loss of at least 1.3 km3 at 25–35 km depth below the point where the four GNSS displace-ment vectors converge, which coincides spatially with the downsag seismicity and whose size is consistent with the variations of the VLP dominant period. A simple hypothesis is that the easternmost edge of the reservoir corresponds with the location of the earliest seismic-ity burst, later seen to migrate upwards. Thus, two questions remain: assuming a 15 km long reservoir, why is volume loss at the western edge and not the centre of the crack, and why is the shrinking hori-zontal rather than vertical? Reconciling the evidence suggest that a dipping sill-shaped reservoir loosing buoyant magma would shrink at its deeper portion, here to the West, and one-sided drainage cou-pled mechanically with the reservoir’s failing roof may explain the observed horizontal shrinking pattern. Alternatively, a more com-plex sill shape, with a vertical westernmost portion, could account for the crustal deformation pattern. Superposition of outward dip-ping thrust faulting are equivalent to a vertical positive CLVD46, which may explain the model derived from the geodetic data.

In conclusion, our analysis suggests that a subhorizontal magma reservoir of up to 15 km in length lies between the newly discov-ered seamount20 and Mayotte. Its exceptional depth of 30 ± 5 km makes it the deepest reservoir whose evacuation has been observed in ground displacement data. Preliminary estimates suggest that more than 3.4 km3 of magma effused at the seafloor20, making it also the largest geophysically monitored submarine eruption to date. The mechanical processes activated offshore Mayotte repre-sent a scaled-up version of a caldera formation process at its nucle-ating, downsag stage43. Phase IV, currently in progress, involves the growth of deep faults at the western edge of the emptying reser-voir. Observations and modelling of reservoir depletion43 suggest

Northwest Southeast

Water

I: Upward magma propagation

II: Lateral magma propagation

III: Depletion

IV: Depletion and sagging

25–35 km depth

VTs (thrust)

VTs (strike-slip)

VLPs

VTs (strike-slip)

Fig. 5 | Summary schematic. A thin, subhorizontal reservoir feeds a dyke propagating first upward (Phase I) and later laterally southward (Phase II). The drainage of the reservoir’s western edge triggers failure of the overlying rock, whose pressure pulses in turn trigger resonance in the reservoir (Phases III and IV).

Period (s)

10

11

12

13

14

15

16

17

18

19

20

20

40

60

80

100

120

Thi

ckne

ss (

m)

140

160

180

200

Length (km)

5 6 7 8 9 10 11 12 13 14 15

V = 2 km3

V = 4 km3

V = 6 km3

V = 8 km3

T =

15.2

s

T =

15.6

s

11.5 12 12.5125

130

135

140

145

∆d = −7.3 m

∆L = −0.2 km

Fig. 6 | Resonance period of a magma-filled crack as a function of crack length and thickness. Thick black lines are periods of 15.2 and 15.6 s that comprise the dominant periods of VLPs analysed here. Contours of reservoir volumes are shown as thin black lines. Zoomed inset shows an example geometry pathway involving first thinning at constant L and then shortening at constant d for the example length of 12 km. L, crack length; d, crack thickness.

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience92

ArticlesNature GeoscieNce

that, should it continue, fault growth will affect a progressively wider area and become shallower. A particular hazard is posed by a scenario where the outward dipping faults reach the ocean floor causing the entire block to collapse abruptly, and new normal faults begin to propagate upward, reaching closer to Mayotte. We estimate the volume evacuation threshold needed to trigger such a collapse, Vmin, based on an equation developed for calderas47:

Vmin ¼10f ρgh2r2

κð1Þ

where f is rock friction coefficient, ρ is average medium density, g is gravitational acceleration, h is reservoir depth, κ is magma bulk modulus, r is reservoir radius. Using f = 0.6, ρ = 2700 Kg m−3, h = 25 km, r = 7.5 km, κ = 25 GPa we obtain a threshold volume of 230 km3, which is ~50 times the volume of the seamount on May 2019. This scenario appears remote at this stage, but critical to monitor any migration of seismicity or change of focal mecha-nisms, as well as better constrain reservoir and faults geometry and crustal properties.

online contentAny methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary informa-tion, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41561-019-0505-5.

Received: 29 May 2019; Accepted: 14 November 2019; Published online: 6 January 2020

References 1. Lemoine, A. et al. The volcano-tectonic crisis of 2018 east of Mayotte,

Comoros Islands. Preprint at https://doi.org/10.31223/osf.io/d46xj (2019). 2. Ohiminato, T. et al. Waveform inversion of very-long-period impulsive

signals associated with magmatic injection beneath Kilauea Volcano, Hawaii. J. Geophys. Res. 103, 839–862 (1998).

3. Kumagai, H. & Chouet, B. A. Acoustic properties of a crack containing magmatic or hydrothermal fluids. J. Geophys. Res. Solid Earth 105, 25493–25512 (2000).

4. Scrutton, R. A. et al. Constraints on the motion of Madagascar with respect to Africa. Mar. Geol. 43, 1–20 (1981).

5. Phethean, J. et al. Madagascar’s escape from Africa: a high-resolution plate reconstruction for the western Somali Basin and implications for supercontinent dispersal. Geochem. Geophys. Geosyst. 17, 5036–5055 (2016).

6. Reeves, C. The position of Madagascar within Gondwana and its movements during Gondwana dispersal. J. Afr. Earth Sci. 94, 45–57 (2014).

7. Geiger, M. et al. Reappraisal of the timing of the breakup of Gondwana based on sedimentological and seismic evidence from the Morondava Basin, Madagascar. J. Afr. Earth Sci. 38, 363–381 (2004).

8. Coffin, M. & Rabinowitz, P. D. Reconstruction of Madagascar and Africa: evidence from the Davie Fracture Zone and western Somali Basin. J. Geophys. Res. 92, 9385–9406 (1987).

9. Jokat, W. et al. Timing and geometry of early Gondwana breakup. J. Geophys. Res. https://doi.org/10.1029/2002JB001802 (2003).

10. Leinweber, V. T. & Jokat, W. The jurassic history of the Africa—Antarctica corridor—new constraints from magnetic data on the conjugate continental margins.Tectonophysics 530, 87–101 (2012).

11. Mahanjane, E. S. A geotectonic history of the northern Mozambique Basin including the Beira High—a contribution for the understanding of its development. Mar. Petrol. Geol. 36, 1–12 (2012).

12. Mahanjane, E. S. The Davie Fracture Zone and adjacent basins in the offshore Mozambique Marginn—a new insights for the hydrocarbon potential. Mar. Petrol. Geol. 57, 561–571 (2014).

13. Emerick, C. & Duncan, R. Age progressive volcanism in the Comores Archipelago, western Indian Ocean and implications for Somali plate tectonics. Earth Planet. Sci. Lett. 60, 415–428 (1982).

14. Nougier, J. et al. The Comores archipelago in the western Indian Ocean: volcanology geochronology and geodynamic setting. J. Afr. Earth Sci. 5, 135–145 (1986).

15. Michon, L. in Active Volcanoes of the Southwest Indian Ocean (eds Bachelery, P., Lénat, J.-F., Di Muro, A. & Michon, L.) 233–244 (Springer, 2016).

16. Audru, J. et al. Major natural hazards in a tropical volcanic island: a review for Mayotte Island, Comores archipelago, Indian Ocean. Eng. Geol. 114, 364–381 (2010).

17. Delvaux, D. & Barth, A. African stress pattern from formal inversion of focal mechanism data. Tectonophysics 482, 105–128 (2010).

18. Stamps, D. S. et al. Geodetic strain rate model for the east African rift system. Sci. Rep. 8, 732 (2018).

19. Cesca, S. et al. Seismic Catalogues of the 2018–2019 Volcano-seismic Crisis Offshore Mayotte (GFZ Data Services, 2019).

20. Feuillet, N. MAYOBS1 Cruise, RV Marion Dufresne (Institut de Physique du Globe de Paris, 2019); https://doi.org/10.17600/18001217

21. Kumagai, H. et al. Magmatic dike resonances inferred from very-long-period seismic signals. Science 299, 2058–2061 (2003).

22. Aster, R. et al. Very long period oscillations of Mount Erebus volcano. J. Geophys. Res. Solid Earth https://doi.org/10.1029/2002JB002101 (2003).

23. Talandier, J. et al. Unusual seismic activity in 2011 and 2013 at the submarine volcano rocard, Society hot spot (French Polynesia). Geophys. Res. Lett. 43, 4247–4254 (2016).

24. Nikkhoo, M. et al. Compound dislocation models (CDMs) for volcano deformation analyses. Geophys. J. Int. 208, 877–894 (2017).

25. Heimann, S. et al. Grond—a probabilistic earthquake source inversion framework. V. 1.0. GFZ Data Services, http://pyrocko.org/grond/docs/current/ (2018).

26. Davis, P. M. Surface deformation due to inflation of an arbitrarily oriented triaxial ellipsoidal cavity in an elastic half-space, with reference to Kilauea Volcano, Hawaii. J. Geophys. Res. Solid Earth 91, 7429–7438 (1986).

27. Uhira, K. & Toriyama, N. Meeting with the May 29, 2015 eruption of Kuchinoerabujima volcano. Bull. Volcanol. Soc. Jpn. 60, 487–490 (2015).

28. Sigmundsson, F. et al. Segmented lateral dyke growth in a rifting event at Bárðarbunga volcanic system, Iceland. Nature 517, 191–195 (2015).

29. White, R. & McCausland, W. Volcano-tectonic earthquakes: a new tool for estimating intrusive volumes and forecasting eruptions. J. Volcanol. Geotherm. Res. 309, 139–155 (2016).

30. Hayashi, Y. & Morita, Y. An image of a magma intrusion process inferred from precise hypocentral migrations of the earthquake swarm east of the Izu peninsula. Geophys. J. Int 153, 159–174 (2003).

31. Passarelli, L. et al. Stress changes, focal mechanisms, and earthquake scaling laws for the 2000 dike at Miyakejima (Japan). J. Geophys. Res. Solid Earth 120, 4130–4145 (2015).

32. Ruch, J. et al. Oblique rift opening revealed by reoccurring magma injection in central Iceland. Nat. Commun. 7, 12352 (2016).

33. Ágústsdóttir, T. et al. Strike‐slip faulting during the 2014 Bárðarbunga‐Holuhraun dike intrusion, central Iceland. Geophys. Res. Lett. 43, 1495–1503 (2016).

34. Wadge, G. The variation of magma discharge during basaltic eruptions. J. Volcanol. Geotherm. Res. 11, 139–168 (1981).

35. Ferrazzini, V. & Aki, K. Slow waves trapped in a fluid-filled infinite crack: implication for volcanic tremor. J. Geophys. Res. 92, 9215–9223 (1987).

36. Chouet, B. Long-period volcano seismicity: its source and use in eruption forecasting. Nature 380, 6572 (1996).

37. Maeda, Y. & Kumagai, H. A generalized equation for the resonance frequencies of a fluid-filled crack. Geophys. J. Int. 209, 192–201 (2017).

38. Kumagai, H. et al. Very-long period seismic signals and caldera formation at Miyake Island, Japan. Science 293, 687–690 (2001).

39. Kobayashi, T. et al. Very long period seismic signals observed before the caldera formation with the 2000 Miyakejima volcanic activity, Japan. J. Geophys. Res. Solid Earth https://doi.org/10.1029/2007JB005557 (2009).

40. Kobayashi, T. et al. Intermittent inflations recorded by broadband seismometers prior to the caldera formation at Miyake-jima volcano in 2000. Earth Planet. Sci. Lett. 357-358, 145–151 (2012).

41. Munekane, H. et al. Mechanisms of step-like tilt changes and very long period seismic signals during the 2000 Miyakejima eruption: insights from kinematic GPS. J. Geophys. Res. Solid Earth 121, 2932–2946 (2016).

42. Fontaine, F. R. et al. Very- and ultra-long-period seismic signals prior to and during caldera formation on La Reunion Island. Sci. Rep. 9, 8068 (2019).

43. Acocella, V. Understanding caldera structure and development: an overview of analogue models compared to natural calderas. Earth-Sci. Rev. 85, 125–160 (2007).

44. Levy, S. et al. Mechanics of fault reactivation before, during, and after the 2015 eruption of axial seamount. Geology 46, 447–450 (2018).

45. Jónsdóttir, K. et al. Evaluating changes of the Bárdarbunga caldera using repeating earthquakes. AGU Fall Meeting Abstracts, 2017, S22C-07 (2017).

46. Cesca, S. & Heimann, S. in Moment Tensor Solutions: A Useful Tool for Seismotectonics (ed. D’Amico, S.) 163–181, (Springer, 2018).

47. Geshi, N. et al. Evaluating volumes for magma chambers and magma withdrawn for caldera collapse. Earth Planet. Sci. Lett. 396, 107–115 (2014).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© The Author(s), under exclusive licence to Springer Nature Limited 2020

NaTuRE GEoSCIENCE | VOL 13 | JANUARy 2020 | 87–93 | www.nature.com/naturegeoscience 93

Articles Nature GeoscieNce

MethodsDetails about the methods used in this paper are given in the Supplementary information.

Data availabilitySeismic data used in this study pertain to networks II48, IU49, GE50, G51, PF52 and RA53 and are available at IRIS (Incorporated Research Institutions for Seismology), GEOFON (GEO-FOrschungsNetz), ORFEUS EIDA (Observatories and Research Facilities for European Seismology—European Integrated Data Archive) and/or the Réseau Sismologique et Géodésique Français (French seismological and geodetic network, RESIF53) web services. Geodetic data are available at the web facilities of the Nevada Geodetic Laboratory, at the University of Nevada, Reno54.

Code availabilityAll codes used in this work are open source. The codes used to generate individual results are available through the contact information from the original publications. Requests for further materials should be directed to S.C.

References 48. Global Seismograph Network - IRIS/IDA (GSN) (Scripps Institution of

Oceanography, International Federation of Digital Seismograph Networks, 1986); https://doi.org/10.7914/SN/I

49. Global Seismograph Network—IRIS/USGS (GSN) (International Federation of Digital Seismograph Networks, 1988); https://doi.org/10.7914/SN/IU

50. The GE Seismic Network (GEOFON Data Centre, 1993); https://doi.org/10.14470/tr560404

51. GEOSCOPE (Institut de Physique du Globe de Paris and Ecole et Observatoire des Sciences de la Terre de Strasbourg, 1982); https://doi.org/10.18715/geoscope.g

52. Piton de la Fournaise Volcano Observatory Network (Reunion Island) (OVPF) (FDSN, accessed 01 Aug 2019); https://www.fdsn.org/networks/detail/PF/

53. RESIF-RAP Accelerometric Permanent Network (RESIF, RESIF-RAP French Accelerometric Network, 1995); https://doi.org/10.15778/RESIF.RA

54. Blewitt, G., Hammond, W. C. & Kreemer, C. Harnessing the GPS data explosion for interdisciplinary science. Eos https://doi.org/10.1029/2018EO104623 (2018).

acknowledgementsThe discovery of the underwater volcano is a result of the MAYOBS 1 campaign that took place from 2 to 19 May, 2019 aboard the Marion Dufresne oceanographic vessel. This campaign was conducted by several French research institutions and laboratories (IPGP/CNRS/BRGM/IFREMER/IPGS) as part of a CNRS-INSU programme20. We thank CNRS-INSU for making public the location and size of the discovered volcano. M.I. thanks H. Sudhaus for her valuable contribution and acknowledges funding by the German Research Foundation DFG through an Emmy Noether Young Researcher Grant (no. 276464525). G.P. is funded by the German Research Foundation DFG project (no. 362440331), a subproject of “SPP 2017: Mountain Building Processes in 4D” (project number no. 313806092). We thank T. James and T. Davis for revising the English text.

author contributionsS.C. coordinated this project, conceived the manuscript and figures, and analysed, modelled and interpreted local accelerometric data. J.L. performed the VLP and array analysis. S.H. and H.R. performed the MT inversion for VT and VLP events. M.I., M.N. and L.P. analysed and modelled deformation data. G.P. assessed the seismic data quality. E.R., F.C. and T.D. contributed to the interpretation of results and discussion section. S.C., E.R., F.C. and T.D. drafted the manuscript. All authors reviewed the manuscript.

Competing interestsThe authors declare no competing interests.

additional informationSupplementary information is available for this paper at https://doi.org/10.1038/s41561-019-0505-5.

Correspondence and requests for materials should be addressed to S.C.

Peer review information Primary Handling Editors: Stefan Lachowycz; Melissa Plail.

Reprints and permissions information is available at www.nature.com/reprints.

NaTuRE GEoSCIENCE | www.nature.com/naturegeoscience